Inspiration
LLMs are being deployed everywhere — apps, agents, automation — but their security is an afterthought. Prompt injection, data leakage, jailbreaking… most developers don’t even realize they’re exposed until it’s too late.
We built Agentic Security to change that — a system that actively scans, tests, and exposes vulnerabilities in LLM-based applications before attackers do.
💡 What it does
Agentic Security is an intelligent vulnerability scanning platform designed specifically for LLM-powered systems.
Aggregates multiple LLM security scanners into one platform Tests apps against prompt injection, jailbreaks, and data leaks Simulates adversarial attacks automatically Generates detailed vulnerability reports Provides actionable fixes and security recommendations Continuously monitors LLM systems for new threats 🛠️ How we built it Frontend: Dashboard for scanning control and report visualization Backend: Orchestrates multiple scanning tools and attack simulations Core System: Agent-based architecture that autonomously tests models LLM Integration: Uses LLMs to generate adversarial prompts and evaluate responses Pipeline: Input App → Attack Simulation → Multi-Scanner Analysis → Risk Scoring → Report Generation ⚔️ Challenges we ran into Designing effective real-world attack simulations Handling inconsistent outputs from different scanners Avoiding false positives in vulnerability detection Making results understandable for non-security developers Keeping up with rapidly evolving LLM threat landscape 🏆 Accomplishments that we're proud of Built a unified platform for LLM security testing Successfully simulated real attack scenarios (prompt injection, jailbreaks) Created an agent-driven automated testing system Delivered actionable security insights, not just raw data 📚 What we learned LLM security is completely different from traditional security Attack surfaces are dynamic and constantly evolving Automation is critical — manual testing doesn’t scale Clear reporting matters more than complex detection 🔮 What's next for Agentic Security Real-time continuous monitoring for deployed LLM apps Integration with CI/CD pipelines for automated security checks Expanding support for multiple LLM providers (OpenAI, Anthropic, open-source models) Building a community-driven vulnerability database Adding enterprise-grade compliance and reporting
Log in or sign up for Devpost to join the conversation.